Protein Engineering for Thermostability through Deep Evolution

Protein engineering is crucial for expanding natural proteins' applications in various fields. Directed evolution, the most common strategy, is usually labor-intensive, costly, and inefficient. We introduce Deep Evolution (DeepEvo), which harnesses the power of cutting-edge deep learning models for protein property identification and sequence generation from an evolutionary perspective. The method achieved a 26% success rate (8 highly thermostable variants in 30 sequences) on a model enzyme experimentally, representing at least 100-fold efficiency improvement compared to conventional directed evolution. Our findings not only contribute to existing knowledge in the field but also pave the way for developing more efficient protein engineering methods.

Here are the training data for the Variant-generator and Thermo-selector in our DeepEvo approach. The parameters of a working Thermo-selector are also provided, along with a sample usage script.